Abstract

Ab initio molecular dynamics simulations permit to explore structure and dynamics of complex systems including the full electronic structure, however they suffer from severe timescale limitations. In the last years machine learning (ML) potentials have permitted to considerably stretch the timescale exploration pushing the ab initio accuracy beyond these limits. In this talk I will present some examples from our recent research activity, where ML potentials, based on ab initio data, have permitted to address problems at the nanosecond scale. As a first example I will discuss the conformational space of adenosine monophosphate (AMP) in explicit solution, while as a second example, I will present the equilibrium structure of the electric double layer at a defective metal/water interface also including the acid dissociation equilibrium. Challenges, limitations and perspectives will be also discussed.